Skip to main content

View Post [edit]

Poster: CR Saikley Date: Mar 22, 2004 8:35am
Forum: petabox Subject: Via Motherboard Benchmarking

The following was created by Dan Aronson to summarize his benchmarking performed at the Archive During February 2004:

The three main configs that were tested were:

via board with PCI 1Gig card(s)
via board with onboard 10/100 ethernet
hp machine with onboard 10/100 ethernet

The purpose of the tests were to see how the machines performed with
heavy network/disk loads

The first set of tests were raw IPerf numbers (only IPerf running) on
the via board with various network
configs.

Next were a set of disk pounding tests (bonnie) for the via and the hp
(primarily to see if there were any real glaring
problems with the via disk performance vs the hp disk performance). The
via board were able to get higher throughput in general from the disks,
however there was much more saturation of the CPU (the latter makes a
lot of sense given the relative performance of the two CPU's).

The next set of tests were disk pounding on the configurations (bonnie)
while doing lots of network IO. As you can see, the via network
bandwidth suffers greatly (presumable because the CPU is swamped by the
disk work. I assume that if we measured contexts switches, the via would
have many more than the hp box)

The wget tests were for:
read -- simulataneous transfer of gigabyte files on multiple disks on
the remote end to /dev/null on the local end, kind of an idealized
version of being limited by the performance of the remote end
write -- simultaneous transfer of data generated by the remote end
(not reading from the disks at all) to multiple disks on the local end,
kind of an idealized version of doing a data load
read/write -- simultaneous transfer of gigabyte files on multiple
disks on the remote end to multiple disks
on the local end

Attachment: Ethernet Perf 03_04_04.xls